How to split stereo audio privately, with no upload
- Step 1Open the tool on a trusted device — Use your own machine. The FFmpeg 8.1 WebAssembly core loads into the page; once loaded, the split runs locally without contacting a server.
- Step 2Confirm the recording is stereo — The tool reports
2chormonoin the file row from a local decode. Only a true 2-channel file (e.g. a two-track deposition with each party on a channel) can be split into separate parties. - Step 3Add the file — it stays on your device — Drag in the stereo
.wav,.mp3,.m4a,.flac,.ogg, or.opus. The file is read into the browser's in-memory filesystem; it is not transmitted anywhere. - Step 4Run the local split — The tool executes
channelsplit=channel_layout=stereoin WASM and produces two mono WAV files. No options, no network round-trip. - Step 5Save both files locally — Download
-1(left) and-2(right) straight to your disk. Nothing was uploaded, so there is no remote copy to clean up. - Step 6Continue the chain locally if needed — Each party's channel can be cleaned with ai-noise-reducer or voice-eq and recombined with audio-merger — all in-browser, all still private.
What stays on your device
Where each piece of data lives during a split. The audio never leaves the browser.
| Item | Location | Sent to a server? |
|---|---|---|
| Your stereo file | Browser in-memory filesystem | No |
| The split (channelsplit) | FFmpeg 8.1 WASM, in the tab | No |
| Output WAV files | Your local Downloads | No |
| Processed-file counter (signed-in) | JAD dashboard stat | Count only — never file content |
| File names / metadata | Stay local | No |
Split behaviour and output
Fixed behaviour from the processing code — no options panel.
| Property | Value |
|---|---|
| Filter | channelsplit=channel_layout=stereo (FFmpeg 8.1 WASM) |
| Input required | Stereo (exactly 2 channels) |
| Outputs | Two mono files — -1 left, -2 right |
| Format | 16-bit PCM WAV, original sample rate |
| Processing | Local, in-browser; no upload |
| Network after load | Not required for the split itself |
Tier limits
Audio family. Duration cap is per file and separate from size.
| Tier | Max size | Max duration | Files/run |
|---|---|---|---|
| Free | 50 MB | 30 min | 1 |
| Pro | 200 MB | 120 min | 10 |
| Pro + Media | 100 GB | Unlimited | 100 |
| Developer | 100 GB | Unlimited | Unlimited |
Cookbook
Privacy-sensitive splitting scenarios. Each party/source ends up in its own mono WAV (left -1, right -2), all processed locally.
Two-track legal deposition, parties separated locally
A deposition recorded with each party on a separate channel. Splitting locally keeps the recording off any server while giving you per-party audio.
Input: deposition.wav (2ch, counsel=L, witness=R) Local split -> deposition-channel.wav-1 (counsel) deposition-channel.wav-2 (witness) Nothing uploaded. No server copy exists.
Confidential client call split for transcription
Splitting each side of a sales/advisory call helps a transcriptionist attribute speakers — and it never leaves your laptop.
Input: client-call.m4a (2ch, advisor=L, client=R) Local split -> two mono WAVs, one per speaker. Transcribe each side separately; audio stayed on-device.
Run it with the network disconnected
Once the page and FFmpeg core are loaded, you can disconnect from the internet and the split still completes — proof the audio isn't being sent anywhere.
1. Load the tool page (online). 2. Disconnect network. 3. Drop file -> split -> download. Works fully offline -> nothing is uploaded by definition.
Embargoed interview, full-quality preservation
A pre-release artist interview under embargo. Lossless WAV output keeps the master quality intact for the eventual edit.
Input: embargo-interview.flac (2ch) Local split -> embargo-interview-channel.wav-1 embargo-interview-channel.wav-2 Lossless PCM WAV; never transmitted.
HR recording handled on a managed laptop
Sensitive HR audio on a corporate device where uploading to external services is prohibited. The in-browser engine needs no install and no upload.
Constraint: no external uploads, no software installs Browser WASM split -> two mono WAVs saved locally. Policy-compliant: nothing left the machine.
Edge cases and what actually happens
Is any audio sent to a server?
No uploadThe file is read into the browser's in-memory filesystem and processed by FFmpeg 8.1 in WebAssembly. The audio is never transmitted to a JAD server or third party. For signed-in users, only a numeric processed-file counter is recorded for dashboard stats — never the file content.
Running with the network disconnected
SupportedAfter the page and the FFmpeg core have loaded, the split executes locally and does not require network access. You can disconnect and still complete the job — a practical demonstration that nothing is being uploaded.
Input is mono
Rejected (requires stereo)A mono recording has a single channel, so there is nothing to separate. The tool throws Channel splitter requires a stereo input. A two-party recording must have been captured as true stereo/dual-mono to be split per party.
Multichannel recording (more than 2 channels)
Rejected (requires stereo)Only exactly 2 channels are accepted; surround or multi-track files are rejected with the stereo-required error. This tool handles stereo dual-party recordings, not multichannel layouts.
Output format can't be changed
By designOutput is always 16-bit PCM WAV. That preserves quality for chain-of-custody and avoids a lossy re-encode. If a smaller deliverable is needed afterwards, convert locally with wav-to-mp3 — still no upload.
Browser tab closed mid-process
Preserved (local only)Because processing is in-page, closing the tab simply cancels the local job — there is no half-uploaded file left on a server, because nothing was ever uploaded. Re-open and re-run.
Very large confidential WAV exceeds Free duration
413 tier limitFree caps duration at 30 minutes per file (and size at 50 MB), enforced locally before processing. Longer recordings need Pro (120 min) or Pro + Media (unlimited duration). Even on higher tiers, the split stays local.
Both parties audible on both output files
By designIf the recording wasn't true dual-mono — both voices were summed into a correlated stereo image — each channel already contains both parties, and the split copies that faithfully. Per-party separation only works when each party owned a channel at record time.
Frequently asked questions
Is my recording uploaded to JAD or anyone else?
No. The split runs entirely in your browser using FFmpeg 8.1 compiled to WebAssembly. Your audio is read into an in-memory filesystem in the tab and processed there — it never reaches a JAD server or a third party.
Is there a temporary server-side copy I need to delete?
No. Because nothing is uploaded, no server-side copy is ever created. There is nothing to retain, expire, or delete after you're done.
Can I prove it isn't uploading?
Yes — load the page online, then disconnect your network and run the split. It still completes locally, which is only possible because the processing is in-browser and the audio is never transmitted.
What does JAD record about my file?
For a signed-in user, a single processed-file counter is incremented for dashboard statistics — a number only, never the file name or content. Anonymous use records nothing about the file.
Is this suitable for legal or HR recordings?
It's designed for exactly those constraints — the audio never leaves your device, output is lossless WAV for quality preservation, and it works on managed laptops with no install. Always confirm it meets your organisation's specific policy.
What format are the output files?
Two mono 16-bit PCM WAV files at the original sample rate — left is -1, right is -2. WAV keeps the master quality intact. Convert locally with wav-to-mp3 if you need a smaller file.
Does it work on a mono recording?
No. A mono file has one channel and can't be split into two parties — you'll see Channel splitter requires a stereo input. Two-party splitting requires the recording to have been captured as true stereo/dual-mono.
Why are both speakers on both output files?
Because the recording wasn't true dual-mono — the voices were already blended into a stereo image, so each channel holds both. The split copies channels faithfully and can't un-mix audio. Use a recorder's two-track/dual-mono mode next time for clean separation.
Can I keep the whole editing chain private too?
Yes. The companion tools — ai-noise-reducer, voice-eq, audio-merger — all run in-browser with no upload, so the entire workflow stays on your device.
What input formats are accepted?
WAV, MP3, M4A/AAC, FLAC, OGG, and Opus, provided the file is stereo. FFmpeg decodes the container locally before splitting.
How large a confidential file can I process?
Free allows 50 MB / 30 minutes per file; Pro 200 MB / 120 minutes; Pro + Media 100 GB with unlimited duration. Limits are checked locally in the browser before processing begins.
What happens if I close the tab while it's running?
The local job is simply cancelled — and since nothing was uploaded, there's no partial file on any server. Re-open the tool and run it again.
Privacy first
Every JAD Audio tool runs entirely in your browser via FFmpeg (WebAssembly) and RNNoise. Your audio files never leave your device — verified by zero outbound network requests during processing.